LogDet question

[James O. McInerney PhD]

Hi all,

I'm putting together a piece of software for LogDet distance measures.  
In the 2nd edition of Molecular Systematics (Hillis et. al., Eds. 
1996), there is the suggestion that LogDet distances have a problem 
when rate variation exists across sites.  As a suggestion the authors 
recommend removing some of the invariant sites (the exact number to be 
decided from multiple runs of a maximum likelihood program and choosing 
the dataset that gives the highest likelihood).  The theory behind this 
approach is to have a dataset where all sites are evolving at 
approximately the same rate.

My question: Say you remove 10 percent of all invariant sites at the 
             outset.  Closely related taxa will still have an awful lot 
             of invariant sites in a pairwise comparison, whilst more 
             distantly related taxa may have very few.  Should the 
             approach be either (a) PAIRWISE removal of a number of 
             invariant sites or (b) removal of a proportion of 
             invariant sites at the outset?

In the original papers the datasets were modified at the outset to 
include only parsimony-informative sites.

Any suggestions welcome,

james

-- 
---
James O. McInerney                  Phone/Voicemail: +44 171 938 9247
Senior Scientific Officer,          email:j.mcinerney at nhm.ac.uk
The Natural History Museum,         
Cromwell Road,
London SW7 5BD
UK                                 
---